Section 5 FTC Act

The US has no comprehensive AI law. Nevertheless, authorities there can take action against companies that use artificial intelligence to deceive customers. This is made possible by a law from 1914: the Federal Trade Commission Act, or FTC Act for short. Its fifth section – Section 5 – broadly prohibits companies from doing two things: engaging in unfair business practices and engaging in deceptive business practices. Enforcement falls to the Federal Trade Commission, the US trade authority, which roughly corresponds to the responsibilities of German competition and consumer protection authorities. Because the wording is so broadly framed, it can be applied to products that didn’t remotely exist in 1914.

Why a law from 1914 affects AI companies

In Europe, the AI Act regulates in detail what AI systems may and may not do. In the US, there is no comparable federal law. Section 5 fills this gap. It doesn’t prescribe technical requirements, but instead focuses on the claims a company makes about its product. Anyone who claims their software reliably identifies applicant talent or diseases must be able to prove it.

For investors and observers, this matters for a second reason. The agency has developed an unusually harsh remedy: so-called algorithmic disgorgement. In such cases, a company must not only destroy unlawfully collected data, but also the models trained on it. A trained model can have cost millions. This threat hits an AI startup harder than a fine.

A common misconception is that Section 5 only applies to American companies. What matters is the market, not the company’s headquarters. A German company that sells its AI software to US customers also falls under it.

Unfair and deceptive: the provision’s two levers

The term “deceptive” is the lesser charge. It applies when a statement misleads a reasonable consumer on a material point. The agency does not need to prove intent. It is enough that the advertising claim is false or that important limitations were withheld. Typical cases: fabricated accuracy figures, a supposedly automated service that in reality has humans behind it, or chatbots pretending to be real people.

“Unfair” is the more serious charge and requires three elements of proof. The practice must cause substantial harm to consumers. The harm must have been reasonably unavoidable for them. And it must not be outweighed by a greater benefit. This covers, for example, systems deployed without adequate testing that systematically disadvantage people.

A proceeding usually ends not in a verdict but in a settlement. The company pays, changes its conduct, and accepts years of oversight. For other companies, these settlements function as a kind of unofficial rulebook. They show where the agency draws the line, without any legislature having passed a law to that effect.

From Rytr to Rite Aid: the provision in the news

Under the label “Operation AI Comply,” the FTC bundled several proceedings in 2024. Among those affected were providers who allegedly sold customers AI-powered online stores with guaranteed income. The agency took action against the writing service Rytr because its tool could mass-produce fabricated product reviews. Another well-known case involved the drugstore chain Rite Aid and its in-store facial recognition, which wrongly flagged customers as shoplifters.

In news coverage, Section 5 therefore comes up almost every time an AI company in the US runs into trouble with authorities. Anyone reading the quarterly reports of major tech corporations will also find the reference there, in the section on regulatory risks. For companies, this means above all one thing: marketing claims about AI capabilities should be provable. “AI washing” – overselling simple software as artificial intelligence – is exactly the case this provision targets.